|
| 1 | + |
| 2 | +# DrivAerNet++ Submission Guidelines |
| 3 | + |
| 4 | +Thank you for your interest in contributing to the DrivAerNet++ leaderboard! This document outlines the process and requirements for submitting your model's results. |
| 5 | + |
| 6 | +## Submission Process |
| 7 | + |
| 8 | +1. Fork the [DrivAerNet repository](https://github.com/Mohamedelrefaie/DrivAerNet) |
| 9 | +2. Evaluate your model using the official train/validation/test splits |
| 10 | +3. Create a new branch for your submission |
| 11 | +4. Add your results and required files |
| 12 | +5. Submit a pull request |
| 13 | + |
| 14 | +## Required Files |
| 15 | + |
| 16 | +Your submission should include: |
| 17 | + |
| 18 | +1. `model_description.md (optional): |
| 19 | + - Model architecture details |
| 20 | + - Implementation specifics |
| 21 | + - Training configuration and hyperparameters |
| 22 | + - Link to paper (if applicable) |
| 23 | + - Link to trained model weights or inference code |
| 24 | + |
| 25 | +2. `test_results.txt`: |
| 26 | + - Complete evaluation metrics on the test set |
| 27 | + - Inference time statistics |
| 28 | + |
| 29 | +## Evaluation Metrics |
| 30 | + |
| 31 | +### 1. Drag Coefficient Prediction |
| 32 | + |
| 33 | +For drag coefficient prediction, the following metrics must be reported: |
| 34 | + |
| 35 | +```python |
| 36 | +# Required metrics: |
| 37 | +- Mean Squared Error (MSE) |
| 38 | +- Mean Absolute Error (MAE) |
| 39 | +- Maximum Absolute Error (Max AE) |
| 40 | +- R² Score |
| 41 | +- Total inference time and samples processed |
| 42 | +``` |
| 43 | + |
| 44 | +Example test output format: |
| 45 | +``` |
| 46 | +Test MSE: 0.000123 |
| 47 | +Test MAE: 0.008976 |
| 48 | +Max MAE: 0.034567 |
| 49 | +Test R²: 0.9876 |
| 50 | +Total inference time: 12.34s for 1200 samples |
| 51 | +``` |
| 52 | + |
| 53 | +### 2. Surface Field and Volumetric Field Prediction |
| 54 | + |
| 55 | +For surface pressure field and volumetric field predictions, the following metrics must be reported: |
| 56 | + |
| 57 | +```python |
| 58 | +# Required metrics: |
| 59 | +- Mean Squared Error (MSE) |
| 60 | +- Mean Absolute Error (MAE) |
| 61 | +- Maximum Absolute Error (Max AE) |
| 62 | +- Relative L1 Error (%) = mean(|prediction - target|_1 / |target|_1) |
| 63 | +- Relative L2 Error (%) = mean(|prediction - target|_2 / |target|_2) |
| 64 | +- Total inference time and samples processed |
| 65 | +``` |
| 66 | + |
| 67 | +Example test output format: |
| 68 | +``` |
| 69 | +Test MSE: 0.000456 |
| 70 | +Test MAE: 0.012345 |
| 71 | +Max AE: 0.078901 |
| 72 | +Relative L2 Error: 2.345678 |
| 73 | +Relative L1 Error: 1.987654 |
| 74 | +Total inference time: 45.67s for 1200 samples |
| 75 | +``` |
| 76 | + |
| 77 | +## Code Requirements |
| 78 | + |
| 79 | +### Test Function Implementation |
| 80 | + |
| 81 | +Your evaluation code should follow this structure: |
| 82 | + |
| 83 | +```python |
| 84 | +def test_model(model, test_dataloader, config): |
| 85 | + """ |
| 86 | + Test the model using the provided test DataLoader and calculate metrics. |
| 87 | + |
| 88 | + Args: |
| 89 | + model: The trained model to be tested |
| 90 | + test_dataloader: DataLoader for the test set |
| 91 | + config: Configuration dictionary containing model settings |
| 92 | + """ |
| 93 | + model.eval() |
| 94 | + with torch.no_grad(): |
| 95 | + # Implement metric calculations as shown in the example code |
| 96 | + # For drag coefficient: |
| 97 | + mse = F.mse_loss(outputs, targets) |
| 98 | + mae = F.l1_loss(outputs, targets) |
| 99 | + r2 = r2_score(all_preds, all_targets) |
| 100 | + |
| 101 | + # For field predictions: |
| 102 | + # Calculate relative errors |
| 103 | + rel_l2 = torch.mean( |
| 104 | + torch.norm(outputs - targets, p=2, dim=-1) / |
| 105 | + torch.norm(targets, p=2, dim=-1) |
| 106 | + ) |
| 107 | + rel_l1 = torch.mean( |
| 108 | + torch.norm(outputs - targets, p=1, dim=-1) / |
| 109 | + torch.norm(targets, p=1, dim=-1) |
| 110 | + ) |
| 111 | +``` |
| 112 | + |
| 113 | +## Submission Checklist |
| 114 | + |
| 115 | +Before submitting your pull request, ensure: |
| 116 | + |
| 117 | +- [ ] All required metrics are calculated and reported |
| 118 | +- [ ] Results are obtained using the official data splits |
| 119 | +- [ ] Model description is complete and clear |
| 120 | +- [ ] Code follows the provided format for metric calculation |
| 121 | +- [ ] All results are reproducible |
| 122 | + |
| 123 | +## Review Process |
| 124 | + |
| 125 | +1. Your submission will be reviewed for completeness |
| 126 | +2. Results will be verified for correctness |
| 127 | +3. Upon approval, your results will be added to the leaderboard |
| 128 | + |
| 129 | +For questions or clarifications, please contact: |
| 130 | +Mohamed Elrefaie (email: mohamed.elrefaie [at] mit [dot] edu) |
0 commit comments